EASy 2 November 2009Artificial Life Lecture 81 Artificial Life lecture 8 (A) Braitenberg vehicles as an introduction to … (B)... Programming simple basic

2 November 2009

Artificial Life Lecture 8

1

EASy

Artificial Life lecture 8Artificial Life lecture 8Artificial Life lecture 8Artificial Life lecture 8

(A) Braitenberg vehicles as an introduction to …

(B) ... Programming simple basic robots in the Robot Lab.

Rationale:- everyone should have some very basic hands-on experience of the problems involved in building a robot that interacts sensibly with its world.Difference between reality and simulation.Then facilities available for those who want to carry on in their own time, or for course projects.

2 November 2009


2

EASy

Braitenberg VehiclesBraitenberg VehiclesBraitenberg VehiclesBraitenberg Vehicles

Braitenberg (1984), Vehicles: Experiments in Synthetic Psychology, MIT Press.

R. Pfeifer & C. Scheier (1999),"Understanding intelligence" MIT Press

Arkin (1998), Behavior-Based Robotics, MIT Press.

2 November 2009


3

EASy

BVBVBVBV

Simplest vehicle has just one sensor and one motor

It can only change its speed according to the stimulus

The +sign means positive excitation

2 November 2009


4

EASy

BVBVBVBV

Crossed positive connections mean that it will turn towards the side with the greater stimulus

-- an aggressor

Contra-lateral excitation

2 November 2009


5

EASy

BVBVBVBV

If the connections are not crossed, then it turns away from excitation – coward

Ipsilateral excitation

2 November 2009


6

EASy

BVBVBVBV

Uncrossed inhibitory connections (note the –sign) give ipsilateral inhibition:

Turns towards the light and slows down as it gets close

Note: there will need to be some constant positive signal to the motors as well, that gets inhibited by the effects of the stimulus

2 November 2009


7

EASy

BVBVBVBV

Contralateral inhibition:

Slows down on seeing light, but then turns away and speeds up again.

Again, some constant positive signal to motors needed also.

2 November 2009


8

EASy

Simulations of AgentsSimulations of AgentsSimulations of AgentsSimulations of Agents

We generally reckon that simple 2-D simulations of agents should ideally be written by YOU in the language of your choice.

If you want 3-D with (fairly) realistic physics, then a Physics Engine like ODE is invaluable (http://www.ode.org/ )

“ODE is an open source, high performance library for simulating rigid body dynamics “ -- classes provided in January.

Or you could build robots for real – and a lot more interesting issues will crop up.

2 November 2009


9

EASy

Theory and PracticeTheory and PracticeTheory and PracticeTheory and Practice

These sets of conceptual ‘wires’ could be implemented in real wires and capacitors and resistors and …

… or you can emulate all this with a computer

2 November 2009


10

EASy

BasicsBasicsBasicsBasics

Body + nervous system

environment

sensors

motors

Many choices, but in this case body = Lego and

Nervous system = C program running on a mini-computer, hooked up to various sensors and motors

2 November 2009


11

EASy

Program is a cheatProgram is a cheatProgram is a cheatProgram is a cheat

The computer program here is a bit of a cheat – it is emulating a nervous system of components passing real values around in real time.

Actually real values to/from motors/sensors are translated into digital signals, and real time is handled by going round a program loop indefinitely (… caution needed)

SEE my new FAQ !! Relates also to GA+CTRNN exercise http://www.informatics.susx.ac.uk/users/inmanh/easy/alife09/TimeSteps.html

2 November 2009


12

EASy

Lego NXTLego NXTLego NXTLego NXT

We previously used our own EASyMind brick in a Lego structure ----- plug in sensors and motors appropriately ---and write a program to load onto it.

But now Lego has improved, we are using the Lego NXT system

2 November 2009


13

EASy

NXTNXTNXTNXT

Up to 3 motors (built-in rotary encoders, measure speed)

Up to 4 sensors:

• Light (emitter+sensor)

• Sonar (partly relates to distance)

• Touch

• Sound (microphone)

2 November 2009


14

EASy

Warnings about sensors/motorsWarnings about sensors/motorsWarnings about sensors/motorsWarnings about sensors/motors

In simulations, magic sensors deliver reliable information about eg distances. Magic motors move the vehicle precise distances.

In the real world, it is nothing like that.

Real sensors are noisy, and react to far more than you might guess.

Calibration of sensors

Wheels slip on the floor, get jammed on bumps.

2 November 2009


15

EASy

Programming NXTProgramming NXTProgramming NXTProgramming NXT

USB port connects to a PC for downloading programs (also has Bluetooth)

Atmel 32-bit ARM processor, (AT91SAM7S256) 256 KB FLASH, 64 KB RAM and runs at 48 MHz.

There is a mode for (Labview-like) graphically programming it (simple, cannot do ANNs) – but we will be using RobotC (http://www.robotc.net/) which is a fairly basic C programming language with pre-written functions for controlling the NXT sensors motors and other peripherals.

2 November 2009


16

EASy

Useful linksUseful linksUseful linksUseful links

Browse these before the Robot Lab class

RobotC www.robotc.net

Our local programming instructions tinyurl.com/587zdx

For more adventurous (build your own hardware) people

Mindstorms.lego.com/Overview/nxtreme.aspx

2 November 2009


17

EASy

What a simple program looks likeWhat a simple program looks likeWhat a simple program looks likeWhat a simple program looks like

int LeftSensorValue;int RightSensorValue;task main(){ while(1)

{LeftSensorValue = SensorValue(LeftSensor);RightSensorValue = SensorValue(RightSensor);motor[LeftMotor] = RightSensorValue;motor[RightMotor] = LeftSensorValue;

}}

2 November 2009


18

EASy

What does it do?What does it do?What does it do?What does it do?

Every time round the loop, it reads in sensor values from where you told the program to look.

and sets the equivalent power levels (suitably scaled) to whatever you have plugged into the Motor Ports

All sorts of stuff is done in the background, hidden from you: A/D conversion, scaling factors, how fast round the loop?

This is an infinite loop while(1)

Get this going first, then modify and extend

2 November 2009


19

EASy

Worries about cheatingWorries about cheatingWorries about cheatingWorries about cheating

Sometimes people forget they are using a program to emulate a ‘nervous system’, and sometimes this breaks down.

while ( 1) {

… … }

How long does this loop take? Does the time vary according to circumstances? If you want something to happen at a particular rate, how do you ensure that?

2 November 2009


20

EASy

Robot lab classesRobot lab classesRobot lab classesRobot lab classes

MONDAY 2nd Nov 16:00 to 18:00 maxGroup B -- Alastair, Vincent, Nick H, Verena, Damien, Matt, Nicholas T, David

TUESDAY 3rd Nov 11:00 to 13:00 maxGroup A -- Dan, Adam, Chris, Paul, Martha, Patrick, Jim, Roberto

PLACE: in the EASy MSc lab, CH1-2R304

2 people to a robot – the rule is that the worst programmer sits at theKeyboard to write the program(s), and the clumsiest person does theRobot assembly

See http://tinyurl.com/587zdx

2 November 2009


21

EASy

GA+CTRNN exerciseGA+CTRNN exerciseGA+CTRNN exerciseGA+CTRNN exercise

This is an optional exercise – but strongly recommended if you may beusing CTRNNs at some stage, for modelling Dynamical Systems – and particularly ‘robot nervous systems’.

See via the course webpage

www.cogs.susx.ac.uk/users/inmanh/easy/alife09/ga_exercise2.html

Hand in your efforts, for feedback.

2 November 2009


22

EASy

Evolve CTRNN to display "Interesting Dynamics" Evolve CTRNN to display "Interesting Dynamics"

In simulation, node activations are calculated forward through time by straightforward time-slicing using Euler integration;

I.e. choose a suitably small time-slice delta_t, use the above equations to calculate for each node how much their values change in time delta_t, and update them all simultaneously

Then repeat many times over, like a slightly jerky movie, though if your delta_t is small enough, hopefully it will be fairly smooth!

2 November 2009


23

EASy

Simulating a Dynamical System using Update StepsSimulating a Dynamical System using Update StepsSimulating a Dynamical System using Update StepsSimulating a Dynamical System using Update Steps

You will be doing this both for the GA+CTRNN exercise AND in the robot lab class.

So many people misunderstand what one is doing with Update Steps, that I have written a special FAQ

www.cogs.susx.ac.uk/users/inmanh/easy/alife09/TimeSteps.html

It is absolutely ESSENTIAL to make the update-steps small-enough,

So that your ‘jerky approximation’ is close enough to the smooth continuous dynamics of what you are simulating.

2 November 2009


24

EASy

Tangents to a Smooth curveTangents to a Smooth curveTangents to a Smooth curveTangents to a Smooth curve

The Blue Line is close-enough to the curve for only a short distance

2 November 2009


25

EASy

Successive approximations …Successive approximations …Successive approximations …Successive approximations …

… will only work if they are close enough

How close is that ?

2 November 2009


26

EASy

The ultimate testThe ultimate testThe ultimate testThe ultimate test

The rigorous test is:

Pick a timestep that by Rule of Thumb (see below) you hope is small enough, run the code, see what result it gives.

Then run the whole program again with a different smaller timestep, 1/10 the size

If the end-result is (almost) the same, you were OK -- stick with it.

If the end-result is significantly different, your original guess was not small enough. Repeat steps 2-4 until you get there

2 November 2009


27

EASy

Rule of Thumb for DSs including CTRNNsRule of Thumb for DSs including CTRNNsRule of Thumb for DSs including CTRNNsRule of Thumb for DSs including CTRNNs

Look at the smallest relevant timescale of the real dynamics of what you are simulating.

WIth CTRNNs, that might be the smallest time-parameter value that any node can have.

Then pick a simulation-update timestep at least one order of magnitude (1/10) smaller than that

Then you should also go through the rigorous test above, to double-check..

www.cogs.susx.ac.uk/users/inmanh/easy/alife09/TimeSteps.html

2 November 2009


28

EASy

Some Public Service AnnouncementsSome Public Service AnnouncementsSome Public Service AnnouncementsSome Public Service Announcements

1. Looking ahead to doing your Alife project

2. (Plagiarism)

3. Standards in Science

4. Bonfire Night

2 November 2009


29

EASy

Alife projectsAlife projectsAlife projectsAlife projects

Next week I shall give a lecture on ‘how to plan an Alife project’….

… and the following week (deadline to be decided) you should be submitting your proposal for your own project, for feedback.

One strongly recommended approach is to take some existing piece of published work (paper, textbook, on web …) and replicate it – then adjust it, tweak it, play with it, extend it.

2 November 2009


30

EASy

PlagiarismPlagiarismPlagiarismPlagiarism

This of course (…like most other work in science…) requires you to distinguish carefully in your report between ideas/code/text that you learnt from others – and new (or new tweaks to) ideas/code/text that you brought to this yourself.

Any piece of work contains both. The ‘Academic Development Course’ should have covered this issue, but you will also see among my FAQs a FAQ page of my own on this.

Go to

http://www.informatics.susx.ac.uk/users/inmanh/FAQ/FAQ.html

And follow link to “Local access only” FAQs

2 November 2009


31

EASy

Science: Honesty, Integrity, ProofScience: Honesty, Integrity, ProofScience: Honesty, Integrity, ProofScience: Honesty, Integrity, Proof

You should be trying to demonstrate something that you believe to be true, that you can lay out clearly and with integrity, using proper standards of proof. If there are gaps in your argument, you will be scrupulous in searching them out, and bringing them to the attention of your readers.

Honesty and integrity should also include maintaining an open mind -- being able, indeed keen, to question one's own beliefs and revise them when evidence and argument point to new conclusions. Honesty includes being scrupulous to give credit to the origins of ideas, and of course no plagiarism; see my relevant FAQ via my homepage.

2 November 2009


32

EASy

Standards of ProofStandards of Proof

Imagine that you are giving the prosecution case against a defendant in a murder trial, with a potential life prison sentence.

The defence lawyer is convinced his client is innocent, and will do everything possible to point out the flaws in your arguments. For a scientific paper, we expect higher standards than the sort of prosecution case that will convince a jury.

2 November 2009


33

EASy

Presenting EvidencePresenting EvidencePresenting EvidencePresenting Evidence

In a court case, every item that fits into the prosecution's argument will be cited, labelled, presented, and cross-examined by the other side. You must meet at least these standards in a scientific paper.

That is why assertions drawn from the literature must be sourced. That is why quotations from Wikipedia are not acceptable, however useful.

Wikipedia might be in tracking down the proper evidential support. In a court case it is not acceptable to say "I heard someone in the pub say he was the murderer" -- you have to bring that person to court and get them to give evidence under cross-examination..

2 November 2009


34

EASy

34

Reminder for Reminder for ThurThursday Nov 5thsday Nov 5thReminder for Reminder for ThurThursday Nov 5thsday Nov 5th

Thu 5th is Bonfire Night

Lewes Bonfire (6am-2am)

Main processions ~7pm on

6 Bonfires from ~10pm

www.lewesbonfirecouncil.org.uk/

Semi-Compulsory

Artificial Life

Exercise !!!!

Documents

EASy 2 November 2009Artificial Life Lecture 81 Artificial Life lecture 8 (A) Braitenberg vehicles as an introduction to … (B)... Programming simple basic